Generative engine optimisation (GEO)

What it means

Generative engine optimisation, or GEO, means preparing a website’s content so that language models and AI assistants find it, understand it and cite it in their answers. More and more people ask an assistant, which assembles an answer from several sources, instead of scanning search results.

GEO builds on classic SEO, since AI search engines often draw on the same indexes. It adds an emphasis on making text easy to extract and quote: clear definitions up front, paragraphs that make sense on their own, concrete facts, figures and dates, and clear sources.

Technical signals are part of it too: structured data that tells machines what is on a page, an llms.txt file summarising the content for language models, allowing AI crawlers in robots.txt and fast pages that can be read without JavaScript.

How it works in practice

Our glossary is built so that assistants can cite it: each term has a short definition in the first sentence, an explanation in separate paragraphs, a last-revised date and DefinedTerm structured data. The llms.txt files offer all the content as plain text in four languages.

For company websites we recommend starting with content that answers customers’ specific questions: what the service costs, how long it takes, which documents are needed, how the options differ. Assistants cite sources that give precise, verifiable answers, not generic marketing copy.

Credibility in the sense of E-E-A-T matters: who wrote the content, what expertise they have, when it was updated and whether the site’s data agree with other sources such as official registers. Conflicting information, such as different prices on different pages, lowers the chance of being cited.

What to watch out for

The most common mistake is trying to “trick” AI with hidden text or bloated, keyword-stuffed content. Language models judge clarity and consistency with other sources, and such tactics are more likely to hurt in classic search as well.

The second risk is blocking AI crawlers in robots.txt without thinking it through. It may protect the content from being used for training, but it also removes the company from assistants’ answers. The decision should be deliberate, and it is worth distinguishing training crawlers from search crawlers.